Who's in the weights?

How the 13 models score one person

Each model's confidence runs from 0 ("never heard of them") to 100 ("sure who they are"). It's not yes-or-no, and it's not one verdict — for the same person, the 13 answers can land anywhere from 0 to 100. Here's one person — a Georgian judoka:

The bars are the 13 models' confidence. Some are sure; several have no idea who he is — pooled together, those make up his strength. Now do this for 291 people.

More looked up → better known, but loosely

Each dot is one of 185 people. Across the bottom: how often they're looked up — their monthly Wikipedia pageviews. Up the side: how well the 13 models know them, their 13 confidences averaged into one score. It climbs — household names sit near the top — but loosely: among the rarely-looked-up, the models know some people surprisingly well and draw a blank on others just as obscure.

Pageviews are on a log scale — each step right means about 10× more monthly views, so the famous and the obscure fit on one chart. Hover any dot for who it is, its score, and a Wikipedia preview; click through to the article.

Some models are far more confident than others

Average each model's confidence across the people we ran, and the averages run from one model that's sure of almost everyone (about 90 out of 100, top) down to one that's blank on almost everyone (about 18, bottom). So how high a person scores depends a lot on which model you ask, not just on who the person is.

Bar length = that model's average confidence across those people, 0–100. Hover a model's name for its size and knowledge cutoff.

But they mostly rank people the same way

Scoring people high and ranking the same people high are two different things — a model can hand out higher numbers across the board but still put the same people on top. So set the overall levels aside and compare orderings: line up each model's people from best-known to least-known. By that test the models match moderately well: a typical pair ranks the people about 0.65 alike, where 1 would mean identical orderings and 0 means no relation at all. So they mostly agree on who is better-known than whom, even where they disagree on the exact scores. The clear exception is the smallest model, whose ordering barely matches the others.

Your line of work barely matters — except for athletes

You might expect some kinds of people to live in the weights more than others. Mostly they don't: split by occupation, recognition is strikingly flat — six of the seven groups sit within a few points of each other, at broadly similar fame. The one clear exception is athletes, who lag the rest. So apart from sport, what you did for a living barely changes whether the models know you.

Average recognition (0–100) per occupation, across the people we ran; their fame is broadly similar across groups, so this isn't just a fame gap. Athletes (orange) stand out.

A shared name costs you a little

Does sharing your name with other notable people hurt? We compared German footballers whose full name is theirs alone on Wikidata to ones whose exact name is shared by several other notable people, matched on fame — so the only real difference is the name: